Subject: Re: Read performance issues with 2.6.0test5?
From: Austin Gonyou <austin@coremetrics.com>
To: Frank Hellmann <frank@opticalart.de>
Cc: freemyer-ml@NorcrossGroup.com, XFS List <linux-xfs@oss.sgi.com>
In-Reply-To: <3F70417E.2050304@opticalart.de>
References: <3F70417E.2050304@opticalart.de>
Content-Type: text/plain
Content-Transfer-Encoding: 7bit
Organization: Coremetrics, Inc.
Message-Id: <1064333276.1939.4.camel@localhost.localdomain>
Mime-Version: 1.0
Date: Tue, 23 Sep 2003 11:07:56 -0500
Sender: linux-xfs-bounce@oss.sgi.com
Errors-to: linux-xfs-bounce@oss.sgi.com
Precedence: bulk
Content-Length: 1863
Lines: 54

On Tue, 2003-09-23 at 07:50, Frank Hellmann wrote:
> Hi!
> 
> Good suggestion and guess what? Performance is bad on the raw device
> as 
> well. So it is not an XFS issue...

I thought this is what you might see. I'd suggest *not* using MD in the
manner you are, and just letting the Infortrend Controller do the
striping. It will be better for you. If you have the updated firmware
3.31T(I think), then you should be able to even chain them together and
do a raid 0 or raid 1 across trays. I've spoken to our current vendor
about doing this should we get another one. Anyway, I'd also suggest
checking your frame size and your Qlogic config setting for modules.conf
using the SANSurfer software. Hope this helps and good luck. I'd like if
you could, to keep me posted off-list about your testing. I'm highly
interested in where this turns for you.

> Sorry for the wrong guess...
> 
>                 Cheers,
>                                 Frank...
> 
> Greg Freemyer wrote:
> > On Mon, 2003-09-22 at 12:37, Austin Gonyou wrote:
> > 
> > 
> >>I agree that these results are odd. Something is amiss here, but
> there
> >>really are many parts to remove before signifying one or the other.
> You
> >>could always format that MD volume you made as EXT3 or Reiser and
> >>perform the same test.
> >>
> > 
> > If the filesystem is of no value, you could also do you dd to the
> raw MD
> > volume.
> > 
> > If that has similar problems then you can forget about xfs as the
> > bottleneck. 
> > 
> > Greg
> 
> -- 
> --------------------------------------------------------------------------
> Frank Hellmann          Optical Art GmbH           Waterloohain 7a
> Digital Cinema          http://www.opticalart.de   22769 Hamburg
> frank@opticalart.de     Tel: ++49 40 5111051       Fax: ++49 40
> 43169199
-- 
Austin Gonyou <austin@coremetrics.com>
Coremetrics, Inc.